Foreign Policy Determinants: Comparing Realist and Domestic-Political Models of Foreign Policy

Transcription

1 Conflict Management and Peace Science, 22: , 2005 Copyright C Peace Science Society (International) ISSN: print / online DOI: / Foreign Policy Determinants: Comparing Realist and Domestic-Political Models of Foreign Policy MARK SOUVA Department of Political Science Florida State University Tallahassee, Florida, USA Systemic realist arguments of foreign policy decision-making suggest that partisan disagreement stops at the water s edge. A domestic-politics model of foreign policy decisionmaking posits that politics does not stop at the water s edge. Extant research on foreign policy voting in the U.S. Congress is consistent with the systemic realist argument. According to this research, partisan voting is less likely to occur on national security, or high-politics issues, than on low-politics issues. I argue that this research suffers from two flaws. First, it does not measure high-politics in accordance with systemic realist thinking. Second, the goal in addressing the water s-edge question is not to learn if a specific variable, such as high-politics, is significant, but to compare competing models. To this end, it is necessary to engage in a three-cornered fight and conduct a nonnested model discrimination test. After creating a new measure of high-politics, I compare a systemic realist model against a domestic-politics model of foreign policy voting in the House of Representatives from The model discrimination test indicates that the domestic-politics model outperforms the systemic realist model. Institutional dynamics and public opinion are more important for understanding foreign policy voting than are more traditional realist variables. Keywords model testing, liberalism, realism, Congress How important is domestic politics in understanding a nation s foreign policy? According to one of the most prominent theories of international relations, structural realism, domestic politics is not very important. For structural realists, the systemic forces of the international power distribution drive a state s foreign policy behavior. Foreign policy, in this view, is primarily a function of the external environment. One prominent scholar has written that even in the post Cold War period, realism is likely to remain the single most useful instrument in our intellectual toolbox (Walt, 1998, 43) for understanding world politics. Nevertheless, in recent years the finding of a democratic peace has led many international relations scholars to conclude that, in fact, domestic politics does play a central role in foreign policy. The debate over the influence of domestic political factors on international relations is important if we are to isolate the primary causal mechanisms for a given class of events, if we are to identify the primary motivations of political leaders, and if we are to accurately analyze crisis bargaining situations. For instance, if structural realism is accurate, then policy makers should emphasize capabilities and the balance of power when negotiating with an I thank Brandon Prins, Bryan Marshall, Bob Jackson, Kevin Clarke, and the CMPS reviewers and editor for many helpful comments. All weaknesses and errors in the manuscript remain my own responsibility. Address correspondence to Mark Souva, Department of Political Science, Florida State University, Bellamy Building, Tallahassee, FL 32306, USA. msouva@fsu.edu 149

2 150 M. Souva adversary. However, if domestic political arguments are more insightful, then policymakers first need to understand the domestic situation in another country and then tailor their policy accordingly. Answering this question about the importance of domestic politics is challenging for two reasons. First, what counts as a domestic-political explanation of foreign policy depends on an implicit contrast to explanations that are not domestic-political (Fearon, 1998, 291). It is not enough to find that a domestic-level variable is statistically significant to conclude that domestic-political explanations trump systemic explanations, for some systemic explanations encompass domestic-level variables. Second, most research has not employed appropriate statistical tests for assessing competing theoretical models. The standard approach for comparing nonnested models is to group all of the variables in a single model and count how many from each explanation are significant. Clarke (2001, 2003) notes that this is an atheoretical way to compare non-nested models and is not conducive to advancing scientific knowledge. Instead, researchers should engage in three-cornered fights (Lakatos, 1970, 115), comparing the rival models against each other and the data, and then assess the competing models using an appropriate test statistic. Importantly, Clarke s research indicates that the evidence against realism... is overstated (Clarke, 2001, 724). Realist models tend to outperform their competitors. One way to evaluate corresponding systemic and domestic-political theories is to ask if politics stops at the water s edge. When one conceives of a state as a unitary and rational actor primarily concerned about its own security, the foreign policy implication is that politics stops at the water s edge, meaning that when it comes to security issues, politicians put partisanship aside. This theoretical perspective has an analog in American politics: Wildavsky s (1966) two-presidencies thesis. Wildavsky argued that presidents are more likely to receive bipartisan support on foreign policy issues because the international environment encourages Congress to defer to the president. If politics stops at the water s edge as structural realist arguments of foreign policy posit, then two presidencies is exactly what we should observe. In recent years, however, the two-presidencies thesis has been challenged. It is argued that there may have been instances of two presidencies prior to the end of the Vietnam War, but that since that time partisanship has become the norm in foreign policy (Meernik, 1993; Prins & Marshall, 2001). Despite these claims, I argue that none of this research has properly evaluated the politics-stops-at-the-water s-edge hypothesis and that to the extent it has evaluated this argument it supports the systemic realist argument. In the next section, I review the international relations and congressional foreign policy literatures. This review suggests two extant and competing models of bipartisan foreign policy voting, a systemic realist model and a domestic politics model. In the third section, I test the competing models by analyzing the behavior of the U.S. House of Representatives. Where previous research compares these theories by examining individual variables, I assess them by comparing the models, through a non-nested model test. The empirical analysis indicates that the domestic politics model outperforms the systemic realist model. If the goal is to understand what drives international relations, it is more fruitful to focus on domestic political factors than on structural power factors. Theoretical Perspectives: Systemic Realism Versus Domestic Politics Fearon (1998) distinguishes between two types of systemic and domestic-political theories. One type of systemic theory (S1) characterizes states as unitary and rational actors. Systemic theories of this type are general and may include unit-level characteristics. A second type of systemic theory (S2) characterizes states as unitary and rational actors but does not employ unit-level attributes to explain interactions between states. This latter type describes Waltz s

3 Foreign Policy Determinants 151 (1979) structural realism. A domestic-political theory (D1) that corresponds to the first type of systemic theory says that states are not unitary actors and that domestic politics leads to suboptimal foreign policy choices. Domestic-political theories (D2) that correspond to the second type of systemic theory also describe states as nonunitary actors and rely on unit-level attributes. 1 The insight of this typology is that it highlights that domestic-political theories of the second type (D2) may resemble systemic theories of the first type (S1) because the latter may include unit-level attributes. As a result, it is not sufficient to find that a domestic variable is statistically significant to conclude that domestic politics matters, for some systemic theories can encompass domestic variables. Instead, the primary distinction between S1 and D2 theories is whether or not the unit-level factors lead to suboptimal foreign policies. A better way, then, to compare systemic and domestic-political models of foreign policy is to place less emphasis on individual variables and more emphasis on how well the models explain a particular dependent variable. With this purpose in mind, I draw on the congressional foreign policy voting literature to elucidate competing systemic realist and domestic-political models of foreign policy partisanship. A Systemic Realist Model of Foreign Policy Voting What motivates states from the systemic realist perspective? Security. 2 Although systemic realism focuses on states as the primary actors in international politics, the microfoundations of the theory posit that any individual under the same circumstances will act in a similar manner: Wherever agents and agencies are coupled by force and competition rather than by authority and law, we expect to find such behaviors and outcomes (Waltz, 1979, 117; emphasis added). The general implication of this theory for foreign policy is that partisan behavior, politics, is rare on foreign policy. 3 The reason for this is twofold. First, any individual will make similar calculations regarding what to do in foreign policy. The necessities of policy arise from the unregulated competition of states (Waltz, 1979, 117). Policy positions, on foreign policy at least, do not originate from parochial concerns. Further, calculation based on these necessities can discover the policies that will best serve a state s interest (Waltz, 1979, 117). Second, internal bickering, or partisanship, presents a divided front to other nations with whom one s home state may be engaged in some sort of dispute. In turn, a divided front is thought to weaken a state s bargaining position and make it more vulnerable to pressure from an adversary; consequently, politicians put partisanship aside when it comes to foreign policy. 4 Gowa (1998, 321), in a study on the use of U.S. military force, makes a similar argument: A state s reputation with respect to its use of military power abroad is a crucial element of its national security. Because reputations can degrade quickly, attempts to manipulate the use of force abroad for short-run goals can inflict large long-term costs. As a result, when it comes to issues of high-politics (national security) there is a tacit partisan truce (Gowa, 1998, 307). The systemic realist argument explains variation in foreign policy bipartisanship with two central concepts: threat and the type of issue at stake. In the presence of an outside threat, individuals who may normally disagree will put aside their partisan differences to unite against a common adversary. This uniting occurs for the following reasons. First, in 1 Forafuller discussion of each of the systemic and domestic-political types, see Fearon (1998). 2 See, for example, Waltz (1979, 85). 3 Gowa (1999) offers a similar argument that systemic theories imply that politics stops at the water s edge. 4 Some formal theory shows that internal opposition may actually strengthen one s bargaining position. On the other hand, Doyle (1986) points out that when it comes to military conflict a unified nation fares much better than a nonunified state. Moreover, what is central here is not the logical accuracy of the systemic position, but the accuracy of the description.

4 152 M. Souva the presence of an outside threat members of Congress recognize both that they have less information than the president and that there is a need for decisive action, so they tend to defer to the president (Wildavsky, 1966). Second, outside threats generate an in-group/outgroup response that fosters a rally-around-the-flag effect. As an external threat becomes more clear, internal coherence increases and foreign policy bipartisanship becomes more likely. One factor that influences the clarity of an outside threat is the structure of the international system. Structural realists argue that a bipolar system is more peaceful than a multipolar system. Bipolarity is more peaceful largely because there is a lower probability of miscalculating the other side s resolve or capabilities. Under bipolarity the threat is clearer, and with better threat perception, states are better able to establish resolve and prevent war from occurring. Bipolarity is more peaceful, then, because it is simpler: Simplicity breeds certainty; certainty bolsters peace (Mearsheimer, 1990, 17). In the structural realist perspective, a bipolar international system is likely to lead to less partisan behavior than is a multipolar system. Throughout the Cold War, Americans believed, because of the clarity of the external threat emanating from the bipolar international system, that partisan political concerns and other divisive issues should be set aside in the interest of national unity in foreign affairs; collapse of the Soviet threat greatly weakened this belief (Gilpin, 2000, 10). After the collapse of the Soviet Union, is it more accurate to think of the international system as unipolar, with the United States as the world s hegemon? 5 If so, how does unipolarity affect threat perception in the dominant state? First, while some realists may view the United States as the world s hegemon, in the systemic realist view the international system is not unipolar (Mearsheimer, 2001, 381). Second, if there was a unipolar system, then it should result in less bipartisanship than we would observe in a bipolar system. Under unipolarity, there is an absence of any credible threat; therefore, the causal mechanism leading to increased bipartisan behavior is not present. Unipolar systems may be more bipartisan than multipolar systems, but both should be less bipartisan than bipolar systems. The annual change in defense expenditures also taps into the clarity of the external threat (see, e.g. Prins & Marshall, 2001). In time of war, for instance, the threat to a nation is clearer than in time of peace. Defense expenditures are also likely to be higher in the former period. Even when a country is not in war, an increase in defense expenditures indicates a greater public concern about security. From the water s-edge perspective, the more concerned a nation is about security in general, the more united it will be on security issues. The second central concept in the realist, water s-edge perspective is the issue at stake. High-politics (national security) issues are expected to produce greater bipartisan voting than low-politics (general foreign policy) issues because of a widely shared view of the national interest, sustained by a sense of patriotism, which tends to mute partisan and ideological differences (McCormick & Wittkopf, 1992, 32). For precisely this reason, previous research has distinguished between different types of foreign policy issues. Meernik (1993) parcels foreign policy votes into three categories: foreign aid, international economic policy, and all other foreign policy issues. Similarly, Prins and Marshall (2001) distinguish 5 Realist theorists that view the international system as unipolar include Charles Krauthammer ( ), The Unipolar Moment, Foreign Affairs 70(1):23 33, and William Wohlforth (1999), The Stability of a Unipolar World, International Security 24(1):5 41. In addition to Mearsheimer (2001), realists that do not view the international system as unipolar include Christopher Layne (1993), The Unipolar Illusion: Why New Great Powers Will Rise, International Security 17(4):5 51, and Kenneth Waltz (1993), The Emerging Structure of International Politics, International Security 18(2):44 79.

5 Foreign Policy Determinants 153 between three types of foreign policy votes: trade, foreign aid, and all other foreign policy issues. Implicit in this categorization is that trade and foreign aid are unique because they represent low-politics issues, while all other foreign policy issues are high politics. Based on this classification, most empirical research examining the role of issues is consistent with the water s-edge argument. For example, foreign aid and trade votes, according to both Meernik (1993) and Prins and Marshall (2001), are more partisan than are other foreign policy votes. McCormick and Wittkopf (1992, 49) also find that high-politics issues tend to elicit somewhat greater levels of bipartisanship. In summary, for both theoretical and empirical reasons we should expect high-politics issues to be less partisan than other foreign policy issues. This distinction between high- and low-politics issues is related to, but distinct from, the two-presidencies hypothesis, which says the president will receive greater support on foreign than on domestic policy from the Congress (Wildavsky, 1966). The difference between the two views is the composition of the category of interest, the test category. For systemic realists, the category of interest is high-politics foreign policy issues. For the two-presidencies argument, the test group is all foreign and defense policy issues. Recent research suggests that the two-presidencies hypothesis no longer holds (Meernik, 1993; Prins & Marshall, 2001), or at most it only holds for Republican presidents (Bond & Fleisher, 1990). From the water s-edge perspective, however, the question is not whether foreign policy in general is less bipartisan; it is whether high-politics issues receive more bipartisan support than low-politics issues, and existing research that parcels out issues seems to support this view. The above arguments lead to the following systemic realist hypotheses on roll call voting in the House of Representatives: Systemic Realism Hypothesis 1: High-politics (national security) foreign policy votes will exhibit less partisanship than other votes. Systemic Realism Hypothesis 2: The ending of the bipolar international system in 1989 will make foreign policy partisanship more likely in the post Cold War period. Systemic Realism Hypothesis 3: As defense spending increases, foreign policy voting is less likely to be partisan. A Domestic-Politics Model of Foreign Policy Voting On the other hand, a domestic-political argument of international relations posits that the primary goal of a politician is to attain or retain office. In a democratic state, office seeking means politicians place primary emphasis on public opinion and domestic welfare. This constituency connection means that politics does not necessarily stop at the water s edge. Although there are a variety of potential domestic-politics models of foreign policy, the congressional foreign policy literature underscores the role of two concepts: public opinion and institutions. From a domestic-political perspective, public opinion is the primary influence on policy, domestic and foreign. A constituency connection, which operates in at least two ways (Miller & Stokes, 1963, 50), allows public opinion to exercise significant leverage over a legislator s actions. First, districts choose their representatives, and in general they elect individuals who share their general views. Second, for electoral reasons members of Congress carefully consider public opinion when deciding how to vote. While politicians may not be slaves to public opinion, Congressmen feel that their individual legislative actions may have considerable impact on the electorate (Miller & Stokes, 1963, 54). The use of economic sanctions is an illustrative example. Economic sanctions seem to be driven

6 154 M. Souva as much by a desire to appease the public as to accomplish any strategic goals (Lindsay, 1986; Kaempfer & Lowenberg, 1992). Further, since the public almost always prefers other options to military force (Jentleson & Britton, 1998), leaders are likely to carefully take the pulse of the state before employing the military. Although the public may be averse to, or at least cautious regarding, the use of military force, it will support it: whenever the American public views foreign policy matters as most important, the previously noted reluctance to use force in the face of a favorable strategic balance will be offset (Ostrom and Job, 1986, 548). In other words, sufficient public support may trump general military considerations, which is a view that could not differ more from that of systemic realism. While politicians may be influenced by public opinion, on its own public opinion does not entail partisan behavior. An additional premise is needed to establish a relationship between public opinion and partisanship, namely that public opinion is often divided into at least two groups. More generally, when public opinion is sufficiently diverse, the electoral impulse encourages politicians, and parties, to emphasize the differences in the public, making partisanship more likely. On the other hand, if there is a strong public consensus, then we should not observe significant partisanship. Institutions also affect foreign policy decision-making by mediating public opinion, and thereby altering behavioral incentives. In terms of congressional voting, as institutional rules increase the constituency connection, they increase the incentive for partisan voting. The introduction of electronic voting into Congress in the early 1970s underscores how institutional changes affect partisanship. Electronic voting made it easier to take more roll call votes, and it ushered in recorded votes. More votes and especially recorded votes increases the pressure on members of Congress from both constituents and special interests to be active and to take stands on issues (Rohde, 1991, 21). Not surprisingly, this incentive to respond to parochial concerns has led to an increase in the use of amendments. From a domestic-political perspective, an institutional change that increases the constituency connection and interest-group pressures should increase partisan activity. More specifically, Iexpect that votes on amendments will be more partisan than other votes. Institutional changes are not only driven by changes in technology. They may also be motivated by changes in public opinion. In the mid-1970s, for example, the House Democratic caucus passed a series of institutional changes in order to better respond to the demands of its new electoral base anchored in civil rights. First, in 1975, Democratic reformers instituted a Subcommittee Bill of Rights. This strengthened subcommittee chairs vis-à-vis their committee chairs and provided the Democratic caucus with final control over committee operations. Second, the Democratic Caucus instituted a secret ballot for the selection of committee chairs. This reform weakened the seniority privilege and made committee chairs more responsive to the party. Third, Democratic reformers strengthened party leaders by giving the Speaker more discretion over bill referral and stronger control over the Rules committee. Coupled with changes in the parties electoral bases that made the parties more internally homogenous and externally heterogeneous, the domestic-political perspective expects these institutional changes to increase partisan activity (Rohde, 1991; Aldrich & Rohde, 2000). Put differently, the House reforms of the 1970s both reflected changes in each party s core constituency and augmented the hearing of these electoral voices in how members of Congress vote. The domestic political perspective expects that the strong institutional reinforcement of the constituency connection will lead to an increase in partisanship on both domestic and foreign policy (Jacobson, 2000: 29). The above arguments lead to the following domestic-political hypotheses: Domestic-Politics Hypothesis 1: Institutional changes that increase constituency ties will make partisan voting more likely to occur.

7 Foreign Policy Determinants 155 Domestic-Politics Hypothesis 2: As public opinion becomes more cleaved, partisanship is more likely to occur. Research Design and Variables To compare the competing models and test the individual hypotheses on partisan voting, I examine roll call votes in the House of Representatives from 1953 to The sample includes all votes, and the time period spans the Cold War and post Cold war periods, affording a test of the relevant hypotheses. The dependent variable is partisanship, which is the absolute value of the difference between the percentage of Democrats voting yea and the percentage of Republicans voting yea. Higher values on this measure indicate more partisanship, with a value of 1 indicating that the two parties voted opposite of each other, and a score of 0 indicating that they voted alike. As both theories expect variation on partisanship, this measure permits an evaluation of which theory better accounts for this variation. 7 The systemic realist argument requires a distinction between high- and low-politics foreign policy issues. Typically, high politics is viewed as relating to national security, yet given the inherent ambiguity of the term security, this definition does not provide much leverage. A more useful conceptualization is that high politics pertains to relationships with strategic rivals and matters of military intervention. Strategic rivals are a nation s primary threats, thus meeting the security criterion. Military interventions are high-politics issues primarily because they invite attacks on the state, or on representatives of the state, but also because they affect how other states view a state s ability to defend itself and its interests. When a state enters into military operations, strategic rivals and threats learn something about its strength. Since military operations may weaken a state, perhaps by depleting its capabilities or through poor performance, emboldening strategic rivals, military interventions may directly affect the security of a state. Do operationalizations of high politics in previous research accord with this conceptualization? Not very well. Past measures conflate high- and low-politics issues. The following is a general list of the types of foreign policy votes in the U.S. Congress over the last fifty years: embassy regulations, foreign aid, human rights, State Department funding, foreign trade, arms control, terrorism and drug trafficking, treaty ratifications, matters related to strategic rivals, and military interventions. 8 Meernik (1993) and Prins and Marshall (2001) include variables in their analyses for votes pertaining to trade and foreign aid. Implicitly, then, they measure high politics as all foreign policy issues outside of trade and foreign aid. While trade and foreign aid fit comfortably into the low-politics domain, it is not the case that they are the only low-politics foreign policy issues. McCormick and Wittkopf (1992) provide a more robust issue distinction. They distinguish among four types of issues: foreign aid, foreign relations, national security, and trade. They also explicitly define the national security and foreign relations categories as high politics. National security votes deal primarily with authorizations and appropriations for defense, while foreign relations votes include U.S. funding commitments to multilateral agencies,... congressional action on various treaties,...statements of U.S. policy on a host of foreign affairs questions,...and ambassadorial appointments (McCormick & Wittkopf, 1992, 30). This measure, then, also conflates high- and low-politics issues. For instance, it is questionable whether researchers 6 I thank David Rohde for the use of his roll call data. 7 The results do not change if I use the dichotomous measure of bipartisanship employed by Meernik (1993) and Prins and Marshall (2001). 8 This list of issues comes from Rohde s (1991) roll call dataset.

8 156 M. Souva should consider defense authorizations and appropriations as high-politics issues, and one would be hard pressed to find an argument in support of ambassadorial appointments and multilateral agency funding as high politics. Who are the United States s strategic rivals, and what issues are related to strategic rivalry? In the period under investigation, the strategic rivals of the United States were the Soviet Union and China. In turn, I classify the issues of containment and communism as high politics. In support of this view of high politics, Ostrom and Job (1986, 544) argue that the most important issues in the minds of U.S. foreign policy makers were anticommunism and containment of the Soviet Union. Overall, I identify the following as high-politics issues: votes related to containment of communism, votes on arms control with the Soviet Union, and votes on the military interventions in Vietnam and Iraq. Lowpolitics issues include votes on embassy regulations, foreign aid, human rights, foreign trade, international treaties, and State Department funding (see Table 1). The systemic realist perspective also says that the external environment should affect the amount of foreign policy partisanship in Congress. During the Cold War, when the United States was locked in what many perceived as a zero-sum conflict with the Soviet Union, foreign policy issues, especially high-politics votes, should have had more bipartisan support than they do in the post Cold War world in which there is less of a security threat to the United States. To capture this effect, I include a dummy variable, post Cold War, which equals 1 for votes between 1990 and I also include an interaction term for high-politics issues in the post Cold War era. When external threats increase, it is likely that defense expenditures increase. At a minimum, an increase in defense expenditures indicates greater attention to security issues, which should result in less partisan behavior. For an additional measure of external threat, then, I calculated the annual change in defense expenditures, in constant dollars. 9 Where the systemic realist argument emphasizes the structure of the international system and the nature of the issue for explaining foreign policy behavior, the domesticpolitical argument emphasizes the importance of the domestic context, especially public opinion and institutional rules. Central to the domestic-political argument is that public opinion affects foreign policy decision-making. However, data matching public opinion with specific votes does not exist, yet previous research provides a useful way to examine the general influence of public opinion. Destler, Gelb, and Lake (1984) argue that prior to the mid-1970s there was less partisanship in foreign policy because there was a consensus in the public on foreign policy issues. Meernik (1993) and Prins and Marshall (2001) also argue, and find, that the Cold War consensus ended with the Vietnam War. Importantly, the Cold War consensus argument is consistent with the domestic-political model, as it posits that as public opinion changes, as the consensus on foreign policy changes, the level of partisan activity changes. The variable consensus, then, serves as a proxy for public opinion on foreign policy. This variable equals one for the period prior to 1975, zero thereafter. Further, in the 1970s, the House of Representatives experienced institutional changes that strengthened the hand of party leaders at the expense of committee chairs. These changes included a weakening of the seniority norm for committee chairs and the subcommittee bill of rights. As these changes were designed to strengthen constituency ties, they should increase the likelihood of partisan voting. The consensus variable, then, also may serve as a proxy for these institutional changes and their effect on foreign policy. To more directly test the effect of institutions on foreign policy behavior, I include a variable for whether the roll call vote was on an amendment or not. As discussed earlier, 9 Defense expenditure data comes from True (2004).

9 Foreign Policy Determinants 157 legislators often introduce amendments to indicate to interest groups and constituents their stance and activity on an issue; as a result, amendments tend to be parochial and more divisive than are other votes. The variable amendment, which takes on a value of one for votes on first- and second-degree amendments, and zero otherwise, provides a test of this argument. Finally, I include a variable for whether or not the president took a position on the roll call vote. From a domestic political perspective, one might expect that the president is more likely to take a position when a vote is in danger of not passing. If the president is taking a position in order to influence the outcome of the vote, then these votes are likely to be more partisan. The variable presidential position equals one if the president takes a position on the roll call vote, zero otherwise. Appendix A summarizes the measures for each variable and the hypotheses for each model. Regression Analyses Table 2 presents the estimates of the rival foreign policy models. First, following standard practice and to facilitate comparison with previous research, I ran a supermodel, that is, a model that includes variables from both the systemic realist and domestic-politics theories. This analysis (column 2, Table 2) provides moderate support for each theory. All of the TABLE 1 Foreign policy issue codings in congressional research Temporal Low-politics High-politics Research domain classification classification McCormick and N/A N/A Wittkopf, 1990 McCormick and Wittkopf, Foreign aid and trade Foreign relations: funding international agencies; treaties; statements about U.S. policy including sanctions, the U.S. role in Central America, and arms control; ambassadorial appointments. National security: defense authorization and appropriation Meernik, Foreign aid and trade All other foreign policy issues Prins and Marshall, Foreign aid and trade All other foreign policy 2001 Author Embassy regulations, foreign aid, foreign trade, human rights, and state department funding issues Issues related to strategic rivals (Soviet Union and China) and issues related to military interventions

10 158 M. Souva TABLE 2 Regression analysis of partisanship in foreign policy, U.S. House of Representatives, Supermodel Domestic politics Systemic realism Variable β (s.e.) model β (s.e.) model β (s.e.) Amendment.086 (.022).092 (.021) Consensus.059 (.033).113 (.036) Presidential position.062 (.013).058 (.014) Post Cold War period.137 (.038).164 (.036) Defense expenditures.006 (.005).003 (.006) High politics.006 (.036).038 (.037) High politics post Cold War.085 (.045).107 (.042) Constant.294 (.032).383 (.021).311 (.021) N p <.05 one-tail test; robust standard errors. domestic-political variables are statistically significant, with their signs in the hypothesized direction. Half of the systemic realist variables are statistically significant. However, I do not find that partisanship is less likely in the House on high-politics issues. It is likely that the inclusion of noncontroversial low-politics issues in previous measures of high politics accounts for this difference. Although the results of the supermodel may be interesting, a central argument of this paper is that one cannot compare the systemic realist and domestic-political explanations by lumping their variables into a single model. Instead, it is necessary to run separate models and conduct a non-nested model test. When tested as designed that is, as a selfcontained argument how does each model perform? In the domestic-politics model, each variable is statistically significant (see column 3, Table 2). In support of the contention that members of Congress offer amendments that have little chance of passing, but that allow members to take positions that are relevant to their constituents or important interest groups, the amendment variable is statistically significant and positive. More broadly, the amendment variable reflects the influence of domestic factors on foreign policy decision-making, and, as expected by the domestic political theory advanced here, it leads to more partisan behavior. The statistical significance of the consensus variable also demonstrates the importance of domestic factors on foreign policy decision-making. Recall that consensus is both a proxy for the foreign policy public opinion consensus prior to the end of the Vietnam War and marks the advent of institutional changes that were expected to lead to more partisan behavior. As hypothesized, votes during the Cold War consensus period were less partisan than were votes after the consensus ended. Turning to the structural realist model, we observe that several variables are statistically significant, though they have signs counter to what systemic realism anticipates. First, the post Cold War era has seen an increase in partisanship. Next, contrary to realist expectations, high-politics votes, in general, are not less partisan than other votes. However, high-politics votes have been less partisan in the post Cold War era. Given that the threat is less clear in the post Cold War era, systemic realism expects the reverse. Finally, the change in defense expenditures is not statistically significant The defense expenditures variable remains statistically insignificant even if it is interacted with high-politics. The interaction term is also not significant.

11 TABLE 3 Non-nested model test, all votes, Foreign Policy Determinants 159 Observed number of positive predictions Expected number of positive predictions Domestic-politics model 11, versus systemic realist model N = p <.05 one-tail test. While the effects of individual variables are noteworthy and interesting, the central focus of this research is the comparison of competing models. Does one model account for the data better than the other, or are they about equal? To answer this question, I conducted Clarke s (2003) pair-signed test of model discrimination. 11 The purpose of this test is to determine which theory out of a group of plausible theories does the best job of explaining a phenomenon (Clarke, 2001, 731). Clarke s test compares nonnested models by examining the predictions of each model. If both models make similar predictions, then they cannot be distinguished. However, if one model is closer to the mark more often than a competitor model, then one can use the underlying number of cases to determine if this difference is statistically significant. If it is significant, then we can conclude that one model performs better than the other model. Clarke s model comparison test involves two steps. First, it is necessary to generate each model s prediction for every data point. In the sample in Table 2, there are 18,389 cases, so there will be that number of predictions. For each case, the test calls for calculating an individual log-likelihood ratio value, which, for a linear model, is a function of the model s fitted values and residuals. 12 If the models have different numbers of independent variables, a degree-of-freedom correction is necessary. After generating the individual log-likelihoods for each model, the second step is to determine if there is a statistically significant difference between the two models. This is done through a paired sign test, which assesses whether the median log-likelihood ratio is statistically different from zero. If the models are equally close to the true specification, half the log-likelihood ratios should be greater than zero and half should be less than zero. If model f is better than model g, more than half the loglikelihood ratios should be greater than zero (Clarke, 2003, 77). In other words, each time a model s prediction is closer to the true value compared to the other model s prediction for that case, it is counted as a positive prediction. Therefore, to determine if one model is better than the other, one wants to know if model 1 generates more, statistically speaking, positive predictions than model 2 (or vice versa). The model comparison test reveals that the domestic-politics model accounts for more variation in the dependent variable and is statistically better than the systemic realist model (see Table 3). If the models are equal, they should each account for about 50% of the cases. However, in about 64% of the cases, the domestic-politics model outperforms the systemic realist model, a far higher value than what one would expect by chance. 13 The statistical significance of the model discrimination test and the finding that the domestic-politics model performs better than the systemic realist model is an important finding, as it runs counter to earlier research. Recall that previous research has found that 11 I thank Kevin Clarke for the Stata code to implement this test. 12 Foralinear dependent variable, one generates an individual log likelihood by calculating the following: log(2 pi sum((residuals(x)) ˆ2)/N))/2 (1/2) ((residuals(x))/sqrt(sum((residuals(x)) ˆ2)/N))ˆ2). 13 For more information on this test statistic, see Clarke (2003).

12 160 M. Souva TABLE 4 Non-nested model test, only foreign and defense policy votes, Observed number of positive predictions Expected number of positive predictions Domestic-politics model versus systemic realist model N = p <.05 one-tail test. changes in the international system and some foreign policy issues affect the likelihood of a partisan vote. From this, one could conclude that the systemic realist argument is at least as convincing as a domestic-politics argument in accounting for foreign policy partisanship. Indeed, the supermodel offers at least partial support for the structural realist argument. But, after comparing the models head-to-head, one finds that the domestic-politics model performs better. The argument here is not that realist variables are insignificant. At least some realist variables are significant. Rather, the argument is that the realist model does not perform as well as the domestic-politics model. In general, bipartisan foreign policy votes are better explained by domestic-political factors than they are by realist factors. Sensitivity Analysis Next, I analyzed the sensitivity of the results to alternative samples, operationalizations, and model specifications. Although the water s-edge argument compares foreign and domestic policy, one might suggest that realist theory is only meant to assess variation within the foreign policy domain. Thus, looking at all votes is an inappropriate sample; one should only examine foreign policy votes. I re-ran the analysis on this sample (see Table 4) and did not find any significant differences. The domestic-politics model still outperforms the realist model. Next, I remeasured the high-politics variable to only include issues related to the Soviet Union, China, and arms control. This reclassification excludes from the high-politics category votes on military interventions, which in essence excludes votes on the Vietnam and Gulf wars. Although the high-politics concept would seem to cover military action against another country, one could argue that the national security of the United States was not threatened by the conflicts in Vietnam and Iraq, and thus issues pertaining to these conflicts should be excluded. This measure of high politics as votes purely related to communism does not affect the model comparison results. Even with the new measure of high politics, the domestic-politics model still outperforms the systemic realist model. 14 Moreover, when high politics is only measured as votes dealing with the Soviet Union, China, and arms control, it is statistically significant (at the.10 level) and positive in the Cold War era. This is the opposite of what realism expects, but is consistent with a domestic-political argument. Finally, I added another interaction term to the systemic realist model. One might conjecture that national security issues are only likely to be less partisan when the president takes a position on the vote. To incorporate this into the model, I added presidential position to the realist model and interacted high politics and presidential position. This interaction term is not statistically significant, though the presidential position variable is significant. Still, even with the addition of these variables to the systemic realist model, the model discrimination test favors the domestic-politics model. 14 This holds for whether the sample is all votes or only foreign policy votes.

13 Conclusion Foreign Policy Determinants 161 This study argues that theories of international relations are also theories of foreign policy. Systemic realism posits that, when it comes to foreign policy, politics stops at the water s edge. Partisan differences may exist, but for the good of the state, politicians put aside their differences on issues of high politics. On the other hand, a theory of foreign policy grounded in domestic politics emphasizes that politicians are motivated by the desire to retain office and that domestic-political factors are the primary influence on their ability to stay in office. In turn, in the context of the United States, electoral and institutional factors affect the behavior of politicians, even on security issues. From a domestic-political perspective, politics does not stop at the water s edge. Although previous research has identified these arguments, it has not appropriately assessed them. The primary weaknesses in previous research are an overly broad specification of what constitutes a high-politics issue and a failure to compare correctly competing models. After conducting a nonnested model discrimination test, I find that a domestic-politics model of foreign policy outperforms a systemic realist model of foreign policy. This does not mean that realism does not provide any insight into international relations. Rather, the conclusion to draw is that we gain more leverage on understanding international relations by focusing on domestic-political factors. This result is robust to alternative operationalizations of the central concepts, alternative model specifications, and samples. The relative success of the domestic-politics model supports other theoretical arguments regarding the way states behave in the international arena. First, it is consistent with the notion of two-level games. If the systemic realist model were to outperform the domesticpolitics model, then the idea of two-level games would be moot. Second, it lends additional support to democratic peace arguments, which underscore the role of domestic-political factors. In summary, the principle motivating factors in international relations are domestic and not structural balance of power. Future research should analyze additional, competing hypotheses from these models. For instance, if exogenous factors affect foreign policy voting, then they should also affect the nature of the foreign policy agenda. Were high-political issues more common during the Cold War? Are international economic policy issues more common in the post Cold War period? On the other hand, a domestic-political perspective suggests that the agenda may vary with the party in power. References Aldrich, J. R., and D. W. Rohde The consequences of party organization in the House: The role of the majority and minority parties in conditional party government. In Polarized politics, eds. J. Bond and R. Fleisher. Washington, DC: CQ Press. Bond, J. R., and R. Fleisher The President in the legislative arena. Chicago: University of Chicago Press. Clarke, K. A Testing nonnested models of international relations: Reevaluating realism. American Journal of Political Science 45(3): Clarke, K. A Nonparametric model discrimination in international relations. Journal of Conflict Resolution 47(1): Destler, I. M., L. Gelb, and A. Lake Our own worst enemy: The unmaking of American foreign policy. New York: Simon and Schuster. Doyle, M Empires. Ithaca, NY: Cornell University Press. Elman, C Horses for courses: Why not neorealist theories of foreign policy? Security Studies 6(1): Fearon, J. D Domestic politics, foreign policy, and theories of international relations. Annual Review of Political Science 1:

14 162 M. Souva Gilpin, R The challenge of global capitalism. Princeton, NJ: Princeton University Press. Gowa, J Politics at the water s edge: Parties, voters, and the use of force abroad. International Organization 52(2): Gowa, J Ballots and bullets: The elusive democratic peace. Princeton, NJ: Princeton University Press. Jacobson, G. C Party polarization in national politics: The electoral connection. In Polarized politics, eds. J. Bond and R. Fleisher. Washington, DC: CQ Press. Jentleson, B. W., and R. L. Britton Still pretty prudent: Post-Cold War American public opinion on the use of military force. Journal of Conflict Resolution 42(4): Kaempfer, W. H., and A. D. Lowenberg International economic sanctions: A public choice perspective. Boulder, CO: Westview. Krauthammer, C The unipolar moment. Foreign Affairs 70(1): Lakatos, I Falsification and the methodology of scientific research programmes. In Criticism and the growth of knowledge, eds. I. Lakatos and A. Musgrave. New York: Cambridge University Press. Layne, C The unipolar illusion: Why new great powers will rise. International Security 17(4): Lindsay, J Trade sanctions as policy instruments: A re-examination. International Studies Quarterly 30: McCormick, J. M., and E. R. Wittkopf Bipartisanship, partisanship, and ideology in congressional-executive foreign policy relations, Journal of Politics 52(4): McCormick, J. M., and E. R. Wittkopf At the water s edge: The effects of party, ideology, and issues on congressional foreign policy voting, American Politics Quarterly 20: Mearsheimer, J Back to the future: Instability in Europe after the Cold War. International Security 15(1): Mearsheimer, J The tragedy of great power politics. New York: Norton. Meernik, J Presidential support in Congress: Conflict and consensus on foreign and defense policy. The Journal of Politics 55(3): Miller, W. E., and D. E. Stokes Constituency influence in Congress. American Political Science Review 57(1): Ostrom, C. W., and B. L. Job The president and the political use of force. American Political Science Review 79: Ostrom, C. W., and D. Simon Promise and performance: A dynamic model of presidential popularity. American Political Science Review 79(2): Prins, B. C Domestic politics and interstate disputes: Examining U.S. MID involvement and reciprocation, International Interactions 26(4): Prins, B. C., and B. W. Marshall Congressional support of the president: A comparison of foreign, defense, and domestic policy decision-making during and after the Cold War. Presidential Studies Quarterly 31(4): Rohde, D. W Parties and leaders in the postreform house. Chicago: University of Chicago Press. True, J Tracking the purposes of government: National government spending, , Walt, S. M International relations: One world, many theories. Foreign Policy Spring Waltz, K Theory of international politics. New York: Random House. Waltz, K The emerging structure of international politics. International Security 18(2): Waltz, K International politics is not foreign policy. Security Studies 6(1): Waltz, K Evaluating theories. American Political Science Review 91(4): Wildavsky, A The two presidencies. Trans-Action 4: Wohlforth, W The stability of a unipolar world. International Security 24(1): 5 41.